Full uACPI implementation

[48cf]

Most uACPI kernel APIs are stubbed out but they are not essential for what we need, which is basically evaluating AML methods to discover PCI root bridges and their resources

[FlyGoat]

Thanks! Build failing on IA32, @mintsuki maybe cc-runtime should provide libatomic symbols?

[mintsuki]

Nah. The main issue here is that we use -march=i386 and that is why this is failing, another commit is incoming to change it to -march=i686, which we shouldn't have a problem with anyways since pretty much all IA32 systems are at least i686 (aka Pentium Pro and newer).

[mintsuki]

@FlyGoat it's ready now!

[FlyGoat]

I do fell like EFI_PCI_ROOT_BRIDGE_IO_PROTOCOL is a better way to get bridge information, bus and mmio ranges are all in Configuration.

I want to avoid evaluating aml if it's possible. I know a handful of platforms, even including Kabylake CRB that are doing wired stuff in AML that will render hardware in to a non-PC compatible state, and the path going back is never validated by vendors so god knows if it will work.

Probably I'm paranoid but I strongly feel this should be optional. I'm happy to merge this with a global switch flag.

[NS-Clone]

which we shouldn't have a problem with anyways since pretty much all IA32 systems are at least i686 (aka Pentium Pro and newer).

and how many pentium pro machines has uefi without csm? 🤣

[48cf]

It seems like EFI_PCI_ROOT_BRIDGE_IO_PROTOCOL provides the same set of information as evaluating _CRS would, but I am wondering how widely supported it is knowing how bad UEFI implementations can be.

I think we could use the protocol with a fallback to fully initializing uACPI and evaluating _CRS in case the protocol is not present, what do you think @FlyGoat?

I'll get back to this PR in a bit if that's okay, I spent most of the time today basically working on PCI scanning + remapping code and we decided to upstream this separately first.

[FlyGoat]

@48cf Yes for sure, you are always welcomed :-)

[mintsuki]

@FlyGoat please take a look now

[48cf]

For the record, I plan to add the code for conditionally fully initializing uACPI when needed in my next PR that will introduce remapping PCI BARs

[FlyGoat]

LGTM, thanks for your contribution!